Overpath

The End of Machine Work

Dermot O'Connor — Fri, 14 Aug 2026 16:10:03 GMT

I have spent my career infatuated with the next thing.

Building and exiting Boxever (acquired by Sitecore), where we helped world-class brands through the first real wave of big-data personalisation and AI. Putting worker training and compliance into people's pockets at HSI Skillko. Now co-founding Overpath. The obsession has not changed.

I have worked through every major inflection point of the last two decades. Cycling through floppy disks to load a single programme. The screech of a 56k modem and the sense that the internet was a new dimension. Watching the cloud turn hardware into air. Mobile putting the world in a pocket.

Every time, the headlines arrived as a mix of utopian promise and dystopian dread. And every time, the same thing happened underneath the noise. The technology recalibrated how we worked and what we valued. It did not remove the human from the picture.

I still love the technology. I love more what it does for us - how it acts as a lever that lets us lift far more than our own weight.

This wave asks a harder question. Is AI another lever, or the first technology that replaces the person holding it?

Work is where identity gets made

Before answering that, it is worth being precise about what is at stake.

There is a modern habit of treating work as a transaction. Hours in, money out. I disagree with that completely. Work is a cornerstone of the human experience. It is where most of us find identity and purpose.

I have always been drawn to Ikigai, the Japanese idea of a reason for being - the intersection of what you love, what you are good at, what the world needs, and what you can be paid for. For nearly everyone I know, the workplace is the main arena where that intersection actually happens.

It is also where we find community. The Harvard Study of Adult Development, the longest-running study of happiness we have, keeps landing on the same finding: relationships are the strongest predictor of long-term health and contentment. A large share of those relationships get forged at work. McKinsey's research on purpose points in the same direction.

Take work away, or strip the personality out of it, and the loss is not measured in productivity. It is the social glue.

Work has given me a mountain to climb and a team to climb it with. Dave Kelliher and Alan Dunne at Datalex gave me the early learnings that shaped a whole trajectory. Dave O'Flanagan and Alan Giles and I spent a decade building and selling Boxever. Brendan Maloney and Liam Brennan and I grew a flourishing tech business organically out of the West of Ireland. Now Ross and I are building Overpath. Those people, and many more I have no room to name, have defined me.

So the conversation about AI and jobs is bigger than tasks. It reaches into the fabric of belonging.

The case that this time is different

I read the clickbait. The end of the white-collar worker.

Then I talk to friends of mine (people I consider significantly smarter than me) and some of them genuinely believe in a jobless future. They point to the speed of displacement and to how well these models now reason and create.

The concern is serious. The World Economic Forum's Future of Jobs work set 69 million new roles against 83 million displaced by 2027. A net loss of 14 million. That number deserves respect rather than a quick counter-argument.

Two decades in this industry taught me a pattern anyway. We overestimate displacement and underestimate evolution.

Think about the mid-nineties panic that electronic mail and digital databases would empty the office. Remove the paper, the filing, the physical hand-offs, and what would be left for people to do?

The internet exploded work. It cleared out the low-value movement of paper so we could absorb a hundredfold increase in information and complexity. It let a small team in Dublin compete with a giant in New York.

Generative AI is the next phase of the same liberation. The fear today is the 1995 fear. The old tasks are visibly disappearing. The complex new world that this efficiency creates is not visible yet.

The end of machine work

For decades we have asked humans to behave like machines. Data entry. Summarising long threads. Keeping rigid CRM fields current. It is workflow debt, and it drains the energy people should be spending on judgement.

That is the work generative AI takes.

The evidence is early and consistent. MIT and Stanford researchers found that access to generative AI raised worker productivity by around 14%, with the largest gains among less experienced workers. Goldman Sachs put 300 million jobs in the zone of exposure and concluded that complement is more likely than replacement, because most roles are only partially exposed. On that reading, keeping a human in the loop is a performance requirement before it is a safety feature.

Generative AI is the first technology capable of saying: I will handle the machine work. Go back to being a human.

From information to execution

Here is the shift I care about most, and the reason I am building what I am building.

The internet connected us to information. AI connects us to execution.

Every organisation I have worked in knew more than it could act on. The playbook existed. The methodology was trained. The data sat in the CRM. And the gap between what the company knew and what any individual did on a Tuesday afternoon stayed open.

Knowledge was never the constraint. The gap is structural, and nothing in the stack was built to close it. The CRM records what happened. Intelligence platforms analyse what it means. Neither one changes what happens next.

That is the layer we are building at Overpath. An execution layer for revenue teams - teammates that carry the deal context, hold the methodology, and show up at the moment the work actually gets done.

The machine work leaves. Judgement, relationships, and the reason people get out of bed stay exactly where they are.

Better boots

The enterprise moves from a run state to a change state. Less time feeding the database. More time interpreting the story the data is telling.

We are in the forming stage of that era and it is messy. The internet, the cloud and mobile all taught me the same thing. Technology expands the theatre we play in.

Work will stay important. We will always need a reason to get up and a group to belong to.

AI isn't the mountain. It's a better pair of boots for the climb.

Why two Sales Reps with the same training convert differently

marketing@overpath.ai (Molly) — Fri, 14 Aug 2026 11:16:33 GMT

Every pipeline review has a moment like this. One rep is converting at 30%. Two others, sitting in the same room, working the same territory, are converting at 12%. They went through the same onboarding. They carry the same playbook. They report to the same manager and work the same tools.

The instinct is to explain this away. Talent, most leaders conclude. Some people are just better at this. But that explanation does not survive contact with the data. The rep converting at 30% was not born knowing how to run a discovery call. They were trained exactly like everyone else in the room, sat through the same certification, and passed it.

What separates them is not what they know. It is what they do, consistently, on every call, without anyone checking. And that distinction is the one revenue leaders have been paid to close for a decade, using tools that were never built to close it.

Talent explains nothing on its own

If talent were the answer, it would show up at hiring - not six months into identical onboarding, identical certification, identical tooling. The 12% reps passed the same assessments the 30% rep did.

They know the methodology. They can name the framework, describe the ideal discovery call, list the signs a deal is drifting.

Talent is the explanation leaders reach for because it requires nothing of the system around the rep. It also explains nothing. A gap that shows up equally between three people with equal training is not a talent gap. It is a gap in what the system allows to happen consistently.

Training tested the wrong variable

The MEDDPICC rollout happened. The workshop ran. The certification was completed, and everyone passed. Ask the middle cohort to describe a strong discovery call and they can do it as fluently as the top performer. The knowledge was tested, and it held.

The gap persisted anyway.

Training measures what people know. It has never measured, and was never built to measure, what they do on a Tuesday afternoon, on the ninth call of the week, with no one in the room and no immediate consequence for skipping the harder question. Knowledge and behaviour are not the same asset.

Training builds the first. It has no mechanism for the second.

Coaching and tools operate on the wrong side of the moment

Manager coaching requires the manager's presence. It scales to the calls they sit in on - a fraction of the total, and rarely the ones that matter most, because managers cannot predict in advance which call will be the one where the deal turns. Intelligence platforms record what happened and analyse what it means, after the call has already ended and the deal has already moved. Both are real investments.

Both improve something real: coaching sharpens judgement over time, intelligence tools surface patterns a manager would otherwise miss. Neither one is present at the moment the behaviour actually happens or doesn't. They sit on either side of the gap. Nothing sits inside it.

The gap is execution

The distinction between the 30% rep and the 12% rep is not upstream, in hiring or training. It is not downstream, in analysis or coaching cadence. It is in the moment itself - the specific question asked or skipped, the stakeholder mapped or left unmapped, the follow-up sent within the hour or three days later once the buyer's attention has already moved on. That moment has never had a layer built for it.

The CRM records what happened. Intelligence platforms analyse what it means.

Nothing has been built to act at the point where execution actually happens, deal by deal, call by call.

This is not a performance gap in the rep. It is a structural gap in the revenue stack, and it will produce the same 30/12 split next quarter, with a different name on the spreadsheet, until something closes it.

Two things to do with this before the next pipeline review

Audit the moment, not the outcome. Pull the transcripts from one comparable stage (the second discovery call, say) for the top rep and the middle cohort, on deals of similar size and complexity. Do not compare quarterly conversion. Compare the single moment: how deep did the pain question go, was the economic buyer named, was the next step scheduled on the call or left for a follow-up email. The gap will usually be visible in one moment, repeated, not in a general lack of skill.
Separate what the team can describe from what the team does. Most enablement measures the first and assumes it produces the second. Ask a sample of reps to describe the ideal handling of a common objection, then compare that description against what they actually said the last time that objection came up on a call. If the two rarely match, the team does not have a knowledge gap to close. It has an execution gap, and no amount of additional training will touch it.

Both of these will surface the same thing: a team that already knows more than it consistently does. That is not a reason to run another workshop. It is the specific evidence a leader needs before deciding what has to change.

The knowledge has never been the problem. The gap is in what happens after the knowing - on the call, in the moment, with no one watching. That gap is execution. It is also, for the first time, solvable.

A Rollout Is an Event. Execution Is Infrastructure

eoin.hamilton@overpath.ai (Eoin Hamilton) — Wed, 22 Jul 2026 12:12:17 GMT

Kickoff lands in week one. Certification is done by week three. The dashboard goes live in week four and for a fortnight the pipeline looks different - cleaner stages, fuller fields, reps using the language from the deck. Then week six arrives and the team is selling the way it sold in April. Most revenue leaders have watched this happen twice. The reason is simple enough to state in one line: a rollout has an end date, and behaviour that depends on an event ends when the event does.

Every lever a revenue leader has comes with a finish date

Look at what a revenue leader actually has to work with.

There is the workshop, which produces two days of energy.

There is the certification programme, which confirms that people can describe the methodology.

There is the enablement launch, the quarterly business review, the weekly one-to-one, and the manager sitting in on the deals they have time to sit in on.

Every one of these has a start date and a finish. Every one produces a spike in behaviour that begins decaying the moment it ends. Each tool does the job it was built for, and the job it was built for is delivering knowledge on a schedule. But look at what happens in the gaps between those events.

The methodology lives in a slide deck and in the leader's memory. The leader is the enforcement mechanism. That has been the condition of revenue transformation for a decade, and it holds regardless of which framework you rolled out.

The decay is a property of the system

Behaviour that has to be re-initiated stops when nobody initiates it. Enforcement runs on the leader's attention, attention is finite, and it is the first thing to go when the quarter tightens.

Two layers already run permanently

Both of them describe. Neither of them enforces.

The CRM records what happened, and it does so continuously - no launch date, no champion, no adoption campaign. Intelligence platforms analyse what it means, with the same permanence and the same indifference to whether anyone is paying attention this week. Both layers were built to describe, and they describe well. The layer that acts at the moment of execution has never existed in the stack. Every deal, every call, whether or not a human initiates it. Overpath is the execution layer.

Four conditions any execution layer has to meet

That word deserves a testable definition, because infrastructure is a claim a lot of software makes. Infrastructure sits at the point of work rather than adjacent to it. It runs without the leader in the room. It has no completion date. And it holds the same standard on the deals nobody is watching as on the ones that surface in the forecast review. Hold any vendor against those four conditions, including this one.

The question this changes

If drift is structural, a better-designed rollout produces a better-looking version of the same curve, and they have already run that experiment. The question stops being how to make this one stick. It becomes: what has to be permanently in place for the methodology to hold without me?

A rollout ends. The execution layer runs. That difference is the whole argument, and it is why the methodology holds.